Precise parallel volumetric comparison of molecular surfaces and electrostatic isopotentials

Abstract

Geometric comparisons of binding sites and their electrostatic properties can identify subtle variations that select different binding partners and subtle similarities that accommodate similar partners. Because subtle features are central for explaining how proteins achieve specificity, algorithmic efficiency and geometric precision are central to algorithmic design. To address these concerns, this paper presents pClay, the first algorithm to perform parallel and arbitrarily precise comparisons of molecular surfaces and electrostatic isopotentials as geometric solids. pClay was presented at the 2019 Workshop on Algorithms in Bioinformatics (WABI 2019) and is described in expanded detail here, especially with regard to the comparison of electrostatic isopotentials. Earlier methods have generally used parallelism to enhance computational throughput, pClay is the first algorithm to use parallelism to make arbitrarily high precision comparisons practical. It is also the first method to demonstrate that high precision comparisons of geometric solids can yield more precise structural inferences than algorithms that use existing standards of precision. One advantage of added precision is that statistical models can be trained with more accurate data. Using structural data from an existing method, a model of steric variations between binding cavities can overlook 53% of authentic steric influences on specificity, whereas a model trained with data from pClay overlooks none. Our results also demonstrate the parallel performance of pClay on both workstation CPUs and a 61-core Xeon Phi. While slower on one core, additional processor cores rapidly outpaced single core performance and existing methods. Based on these results, it is clear that pClay has applications in the automatic explanation of binding mechanisms and in the rational design of protein binding preferences.

An efficient algorithm calculating common solvent accessible volume

“…To decide the 52 protein structures, we initially selected the 100 protein structures listed in the dataset used by Georgiev et al . [ 26 ]. The dataset includes proteins whose sequences are not redundant with a BLAST p-value cutoff of 10 −7 , which are selected arbitrarily using the VAST server.…”

“…There are many methods designed to measure the solvent accessible surface/volume of a protein [ 11 – 26 ], which are often used in implicit solvent models to describe the forces exerted on atoms on the protein surface by the solvent [ 1 – 7 ]. However, the existing methods have limitations in solving the above integration problem in Eq (5) .…”

“…Many methods have been developed to determine the protein surface area, the protein volume, and the solvent accessible volume [ 11 – 26 ]. Those methods are developed by using Voronoi diagrams [ 11 – 13 ], polyhedrons [ 14 ], the inclusion-exclusion principle [ 15 – 17 ] by finding the overlapping volume of three spheres [ 18 , 19 ], the integration of the solvent accessible surface area [ 20 , 21 ], or the voxels (or cubes) [ 22 – 26 ]. The methods determining the solvent accessible volume (SAV) have been used for the free energy calculation [ 12 , 16 , 17 , 20 , 27 , 28 ].…”

“…Our results show that the proposed method determines the CSAV value in O ( mn log n ), where n is the number of atoms involved in the CSAV calculation and m is the resolution that controls the trade-off between the computational cost and the accuracy. We compare the proposed method with a naïve voxel-based method, which is related to other voxel-based methods [ 22 – 26 ] but is designed to determine the CSAV value. The results show that the proposed method is superior to the voxel-based method in both computational efficiency and accuracy.…”

HBcompare: Classifying Ligand Binding Preferences with Hydrogen Bond Topology

“…HBondFinder defines the edges of the graph by finding all donor-acceptor pairs that satisfy our hydrogen bond criteria, which are inspired by the HBPlus program [ 24 ]. This process is accelerated with a lattice-based geometric data structure [ 25 ] that allows us to rapidly search for all atoms of a specific identity that are within a radius of a given point. This search allows us to find all combinations of the four critical atoms of a hydrogen bond: "D", the hydrogen bond donor, "H", the donor hydrogen, "A", the acceptor, and "AA", the acceptor antecedent.…”